Palo Alto CEO: OpenAI's GPT-5.6 "Good Start," Needs Cost Cuts

Palo Alto Networks CEO Nikesh Arora says OpenAI's new GPT-5.6 Sol model, which is 54% more token-efficient, is "a good start." However, he warns that AI adoption at scale requires much deeper cost reductions. Arora believes token efficiency must drop to 20% within the next year and 90% by the year after that. He states that high token costs are straining enterprise budgets and making AI tools hard to implement. Arora is not alone. Palantir CEO Alex Karp also criticized the token model, suggesting open-weight models could be a solution. Many businesses are now experimenting with cheaper open-weight tools, including models from China. Despite these concerns, AI infrastructure spending is surging. SpaceX recently raised $25 billion, and Amazon raised another $25 billion for AI expansion. Arora notes that while demand for AI is "infinite," budgets will eventually rationalize with more efficient technology. He also recently warned CEOs that layoffs are not the solution to the industry's talent shortage. Arora advises a gradual, specialized hiring strategy. He says 90% of employees at large companies lack AI fluency, a gap he believes is critical for career survival. This comes as a 2025 study found 39% of employers have already made redundancies due to AI. This all matters because the cost and accessibility of AI will shape its widespread use and impact on the workforce.